TensorRT PTQ hands on

You can run the mnist_int8.ipynb to finish this hands-on test. Before opening the notebook, you need to start a docker and do some port mapping as following.

1. Run docker run --rm -it --gpus '"device=0"' -v /your_path:/workspace/hostdir -p xxxx:8888 nvcr.io/nvidia/tensorrt:21.09-py3. /your_path is the dir you want to mount to docker, xxxx is the server port you will use to visit the notebook, and 8888 is the jupyter notebook port within the docker.
2. In docker, please run pip install jupyter && jupyter notebook --ip=0.0.0.0 --NotebookApp.token='' --no-browser --port=8888 --allow-root
3. Tap Win+R and run ssh -N -f -L localhost:xxxx:localhost:xxxx -p 22 [email protected]. xxxx is the port you used in step 1, user_ID is you ID to access to the server, and x.x.x.xis the server IP address.
4. In your host browser, please input http://localhost:8889/

Now you can open the mnist_int8.ipynb file. Please run every cell of this notebook. Then, please finish the questions and TODO tasks in this notebook.